Neotoma
Paleoecological
Database


http://neotomadb.org

1 Introduction

This workshop is designed to use paleoecological data and associated internet resources to provide participants with the following:

  1. Guidance on best practices in the archiving and analysis of paleovertebrate data

  2. Training in the use of the Neotoma Paleoecology Database (www.neotomadb.org) to archive, access, and analyze paleoecological data.

Neotoma is a multiproxy paleo-database that stores a range of paleoecological & paleoenvironmental data, including vertebrate faunal data. One of the strengths of Neotoma is the ability to compare faunal data with other proxy data such as fossil pollen, diatoms, ostracodes, insects, charcoal, and geochemical data. In addition, the database is structured to relate absolute dates to taxon occurrences and to allow the creation and storage of age models built on absolute dates from stratigraphic sections. Neotoma is a public-access, community-supported database that is emerging as the standard repository for Pliocene and Quaternary paleoecological data.

More teaching materials can be found in Neotoma’s educational resources.

2 Finding Data

2.1 Explorer

2.1.1 Getting Started

This is what you see when you’re in the right place.
  1. Go to http://www.neotomadb.org/ and click on the ‘Explorer’ picture, or navigate directly to the Explorer App
  2. Pan (by dragging), or change the zoom so that your window is centered on North America, including all of the lower 48 states and the southern half of Canada.

2.1.2 Tips and tricks

  1. Show/Hide Search Results
    1. You often accumulate many search layers - it may be getting a bit confusing. Find the icon that lets you show/hide/combine search layers and use it to hide or delete some of your searches.
  2. Rename searches
    1. Searches from the “Basic” search window are automatically named, but not “Advanced” searches. Add names for your searches as you go at the bottom of the Search dialog

2.1.3 Search for Data

2.1.3.1 Finding sites

  1. Find a known site
    1. Using the Search dialog window (Advanced tab, Metadata subtab, Site Name field), find the classic Guilday site “New Paris #4”.
      1. Note: Search doesn’t like the # symbol, so search for “New Paris”, which should return two sites: New Paris #2, and New Paris #4
    2. Once you’ve performed the search, click on the point that appears. A window will pop up with some information about this record.

QUESTION 1: What is the latitude and longitude of New Paris #4? What is the Site ID?

  1. Explore sites by geography
    1. Using the Search dialog window (Advanced tab), first choose “dataset type = vertebrate fauna” at the top
    2. Then, in the Space subtab, click “Search by extent”
    3. Click the “Extent” dropdown menu and search by shape, select the rectangle, then draw a rectangle on the map in your chosen region of North America

QUESTION 2: How many total sites are found in that region?

  1. Find all sites produced by a researcher
    1. Search Window, Advanced Tab, Metadata subtab, Person Name field
    2. Find all sites produced by Ernie Lundelius

QUESTION 3: In which states has Ernie worked?

2.1.3.2 Find a Taxon

Help me find my friends!
[Image credit: Wikimedia Foundation]
  1. Search for a single taxon.
    1. Search Window, Basic Tab, use Taxon field
    2. Find all sites with Antilocapra records

QUESTION 4: How many sites have Antilocapra records?

  1. Search for multiple taxa
    1. The basic Antilocapra search indicates that we might need to build up the taxonomy. Let’s re-search for Antilocapra, but make sure we’ve included all relevant records.
    2. Search Window, Advanced tab, Taxa subtab
      1. Note: To do this, we will search for a “Taxa group”.To the right of the “Taxon name” field, click on the gear symbol. Click on “Mammals” for the Taxa group, then search for Antilocapra. Then, click on all taxa you want to include here. For example: Antilocapra americana, Antilocapra sp., Antilocapra, cf. Antilocapra americana, cf Antilocapra sp., Antilocapra cf. A. americana, Antilocapridae ?Antilocapra sp., and ?Antilocapra sp.
  1. Add the modern species range onto the search: Click on the red polygon at the top, search for and select Antilocapra americana

QUESTION 5: What state has the easternmost location of Antilocapra in this search? (For comparison, the eastern range limit of Antilocapra today is in western Iowa).

2.1.3.3 Find all vertebrate records in North America

  1. Search window, Advanced Tab, Dataset type = “Vertebrate fauna”

QUESTION 6: The generation of fossil records is labor intensive and hence expensive - e.g. the costs of fieldwork, the money spent on radiocarbon dates, the time required for a trained analysis to identify specimens, etc. A rough time/cost estimate for a single vertebrate fossil record is on the order of two years and $30,000. Given this, give an order-of-magnitude estimate of the number of person-years and dollars it took to generate these fossil vertebrate records now stored in Neotoma. (Order-of-magnitude = 10 person-years? 100 person-years? 1000? etc.)

2.1.4 View Data

  1. Using the last Mammuthus search (between 11,000 - 8,000 years ago), click on the single site that appears (Fetterman Mammoth Locality).
  2. In the popup window with metadata for Fetterman, note that the bottom includes a list of datasets available at the site. (V = vertebrate; clock = geochronological data)
  3. Hold the mouse over the vertebrate dataset. Note that an eyeball and ‘+’ appear to the right.
  4. Click on the eyeball to view the Fetterman Mammoth Locality vertebrate dataset. This opens up a new window with more detail about your dataset.
    1. Samples: A data table. Each row is a different variable, the first few columns show metadata associated with each specimen, and each subsequent column is a stratigraphic depth. In this case, this is an assemblage so there is only one depth column.
      1. Explore the specimens found at the site.
    2. Site: General information about the site
    3. Chronology: There is no associated geochronological dataset, but the chronology tab indicates an age for the mammoth.

QUESTION 9: What other information would you want about this date to determine authenticity?


QUESTION 10: What publications are listed for Fetterman?

Well. . . Play it again Sam!
[Image credit: UCMP.]
  1. Now search for the site “Samwell Cave”.
    1. Notice that this site has multiple localities within it, and each locality has both a vertebrate dataset and a geochronological dataset.
    2. Explore the Samwell Cave Popcorn Dome datasets further.

QUESTION 11: What is the difference between information in the “Chronology” tab of the Vertebrate dataset versus the information within the Geochronology dataset?

2.1.5 Download Data

  1. In the popup window with metadata for Samwell Cave Popcorn Dome, hold the mouse over the vertebrate dataset. Note that an eyeball and ‘+’ appear to the right. Click on the ‘+’. This adds the dataset to a ‘Datasets’ tray.
  2. Find the icon for the ‘Saved Datasets Tray’ and click on it.
  3. Hover over the dataset then click on the Save icon to the right. The dataset will be saved as a text file in CSV (comma separated value) format.
  4. Open the downloaded CSV file in Excel or a text editor (e.g. Notepad, Wordpad) to look at it.

QUESTION 12: When was the last time Aplodontia rufa was seen at the site?

3 Web API- Application Programming Interface

3.1 What is a Web API?

  1. Structured and documented way of communicating with web services and incorporating them in your own application.
  2. Often used to access a database, such as Neotoma, iDigBio, or the Paleobiology Database.
  3. Calls are usually sent as URL strings: values appended to a URL after a ? are processed as parameters to be passed to the server. Most APIs use calls to different URLs to return different kinds of data or the specific data that meets your query. We can think of http://google.com/?q=bananas as if it were an R function called search(), passed as search(q="bananas").
    • URL = Uniform Resource Locator – address of a web page.
    • URI = Uniform Resource Identifier – web address of a service, like an API.

Many public APIs exist. For example, Google has APIs for most of its products. You could write an R (or Python, etc.) script that would connect to the Google Calendar API to allow you to automatically change events on your calendar or report agenda items back to you in your own custom environment. Hilary Parker has a nice example for adding sunsets to Google Calendar using R.

APIs are an important part of many important biological and paleobiological databases, because they allow users to filter, search, and download records using custom scripts. Different APIs have different parameter names, structures, and can produce different types of returns, including raw text, comma-delimited test, and JSON (JavaScript Object Notation).

We will work through examples from the Paleobiology Database and Neotoma.

3.2 Example 1: Text returns from the Paleobiology database.

Goal: Get all occurrences of Camelidae in the Paleobiology database that are dated to the Pliestocene, returning only the spatial location of the occurrence. Get the response as comma-separated text (CSV format).

  1. The URI for the occurrences returned in text format is https://paleobiodb.org/data1.2/occs/list.txt.
  2. To specify parameters, add a ?, followed by the parameter key-value pairs. base_name=Camelidae& searches for a taxonomic name at any level of the Linnaean hierarchy and returns any occurrences with that name, its synonyms, or its subtaxa. Try changing to a name of interest to you.
  3. To combine multiple search or filter parameters in a single request a request, use a & between key-value pairs. &interval=Pleistocene returns occurrences that fall within the specific geologic interval. Intervals can include North American Land Mammal Ages as well as the conventional Geologic Timescale. Try “Blancan” and “Burdigalian” to see what happens.
  4. Finally, the PaleoBioDB API allows you to pick what data gets returned with your call using the show parameter. show=loc limits the data of the response to only include the location of the occurrences.

You should end up with a URI string like this, which can be typed into your web browser:

https://paleobiodb.org/data1.2/occs/list.txt?base_name=Camelidae&interval=Pleistocene&show=loc

There are many useful parameters to facilitate searching occurreneces in the Paleobiology database, including:

  • taxonomy
  • age
  • environment
  • location
  • geographic bounding box (lngmin and lngmax as well as latmin and latmax)
  • database ID

The list of potential parameters is here: https://paleobiodb.org/data1.2/occs/list_doc.html.

Try for a few minutes to explore the data with different calls, using the call list.

The full documentation for the Paelobiology Database API is located at https://paleobiodb.org/data1.2/

3.3 JSON

Text returns are easily read by humans, but are often not suitable for the nested data structure saved in relational databases. JavaScript Object Notation (JSON) is a widely used data format that can encode these hierarchical relationships for sending across the Web, is easy to read, and can be consumed by many scripting languages (javascript, R, python, etc). JSON is a newer alternative to XML (eXtensible Markup Language), which you may have seen before. JSON is becoming the standard for data transfer in web services and R has several packages for dealing with JSON-formatted data.

JSON is composed of key-value pairs, enclosed by curly brackets and separated by commas (objects). You may also present an array, or an ordered collection of values, enclosed in square brackets. Values may be their own objects, allowing nested relationships between data.

As an example, you could present an occurrence like this example from the Neotoma API:

{
  "SiteLongitudeWest": -103.31666666666666,
  "SiteLatitudeSouth": 34.283333333333339,
  "TaxonName": "Smilodon fatalis",
  "VariableElement": "bone/tooth",
  "Value": 1.0,
  "VariableContext": null,
  "TaxaGroup": "MAM",
  "SampleAgeYounger": 15332.0,
  "SampleAgeOlder": 30041.0,
  "SiteLongitudeEast": -103.31666666666666,
  "SiteAltitude": 1280.0,
  "VariableUnits": "present/absent",
  "DatasetID": 4564,
  "SampleAge": null,
  "SiteLatitudeNorth": 34.283333333333339
}

Can you tell what kind of occurrence this JSON object is describing? How old is it? Where is it located?

3.4 Example 2: Neotoma, JSON, and R

Goal: Search Neotoma for all Smilodon occurrences, and put them into an R data frame.

  1. The Neotoma database has a different structure and requires different parameters than the Paleobiology database. Here is the example API call:

http://api.neotomadb.org/v1/data/sampledata?taxonname=Smilodon*

  1. Notice when you make this call that the JSON is computer-friendly but not human-friendly. Try it again with the ‘pretty’ format tag:

http://api.neotomadb.org/v1/data/sampledata?taxonname=Smilodon*&format=pretty

You should be able to see the nested set of JSON objects, including the occurrences returned as comma-separated objects within "data".

Try experimenting with the search, substituting different names. We’ll do more with the Neotoma API in a bit.

JSON is becoming the standard for data transfer in web services. R has several packages for dealing with JSON-formatted data. We will use some examples from the package RJSONIO. We will also use the package RCurl, which has functions to let you query APIs from within the R environment. The neotoma package is built around the newer httr package.

# Uncomment this line if you haven't already installed any of these packages:
# install.packages(c("RCurl", "RJSONIO"))

# Add the packages to your programming environment 
library(RCurl)
library(RJSONIO)

You will also want to make sure you have the latest version of R installed, or else the secure connection (https) won’t work in the following queries.

Now create a query for the PaleoBioDB API using the example from above:

q <- "https://paleobiodb.org/data1.2/occs/list.txt?base_name=Camelidae&interval=Pleistocene&show=loc,class"

Create an object to receive the results:

a <- basicTextGatherer()

And execute that query:

curlPerform(url = q, writefunction = a$update)

Finally, view the data:

a$value()

You can see that the data have come in as a character vector; a long list of text strings with no clear structure. Luckily, the PaleoBioDB API also has a JSON interface.

Change your query to refer to list.json:

q <- "https://paleobiodb.org/data1.2/occs/list.json?base_name=Camelidae&interval=Pleistocene&show=loc,class"

Rerun your query and look at a$value( ) again. How does it look now?

a <- basicTextGatherer()
curlPerform(url = q, writefunction = a$update)
a$value()

You can convert it to a data frame:

tmp <- fromJSON(a$value())
records <- tmp$records
results <- data.frame(records[1], stringsAsFactors = FALSE)
for (x in records[-1]) {
  x <- data.frame(x, stringsAsFactors = FALSE)
  results <- merge(results, x, by = intersect(names(results),names(x)), all = TRUE)
}

dim(results)  # this shows the dimensions of the data frame
head(results)  # using head() here only prints the top 6 rows, out of 424

When you have gotten this code to work, go through and make comments to describe what each section is doing. Remember, you can make comments in your R code by placing a hashtag (#) at the beginning of a line, or after a line has run (as with the last few lines in the above chunk of code).

Try playing with your queries to see what you can pull from the PaleoBioDB, or even extending to the other two APIs we have explored.

3.5 R packages and Database APIs

In the previous example, you can see that the example API call is reporting data from only part of the distributed database schema of Neotoma. In fact, the Neotoma API is designed around a set of different URLs, each of which allows a user to search a portion of the database. So, if you want the site information for occurrences of Smilodon, you would have to search on Smilodon in the SampleData URI (as we did in the example), pull the DatasetID values from those returns, then search on the Dataset ID (http://api.neotomadb.org/v1/data/datasets) for those DatasetIDs, which would, in turn, produce the SiteIDs, which you would then search on the Sites URI (http://api.neotomadb.org/v1/data/sites). This sort of searching would be cumbersome if you were to do it by hand, but fortunately you can script a computer to do it for you. In fact, you don’t have to write the scripts to do it, because they have already been constructed and provided to the community as the R neotoma package.

Other APIs also have wrapper packages to simplify data calls in R. The PaleoBioDB has a package, paleobioDB, and iDigBio has a package, ridigbio. Currently, a large group of collaborators is working on a single API and wrapping R package to access both PaleoBioDB and Neotoma at the same time, as well as linking to many online museum databases and iDigBio. While these packages are a great convenience and the neotoma package is the subject of our next module, we wanted to introduce you to the underlying architecture here so you would understand what these packages are doing. If they cannot give you the data or fomat you need for your work, they can be cracked open and a custom solution can be easily put together.

4 The neotoma Package

Install the neotoma package, then add it to your programming environment

# Uncomment this line if you haven't already installed any of these packages:
# install.packages(c("neotoma"))

# Note, because the `neotoma` package is under active development there may be very recent changes
# you may be interested in.  To get those, uncomment the following lines:
# install.packages('devtools')
# devtools::install_github('ropensci/neotoma')

library(neotoma)

neotoma has three core commands: get_site, get_dataset, and get_download. The first two return metadata for sites and datasets; the latter returns data. See Goring et al. (Goring et al. 2015) for a full description of the package and example code. This exercise is partially based on those examples.

4.0.1 Finding sites

We’ll start with get_site. get_site returns a data.frame with metadata about sites. You can use this to find the spatial coverage of data in a region (using get_site with a bounding box), or to get explicit site information easily from more complex data objects. Use the command ?get_site to see all the options available.

You can easily search by site name, for example, finding “Samwell Cave”.

samwell_site <- get_site(sitename = 'Samwell%')

Examine the results

print(samwell_site)
##     site.name      long      lat elev
##  Samwell Cave -122.2379 40.91691  465
## A site object containing 1 sites and 8 parameters.

While samwell_site is a data.frame it also has class site, that’s why the print output looks a little different than a standard data.frame. That also allows you to use some of the other neotoma functions more easily.

By default the search string is explicit, but because older sites, especially pollen sites entered as part of COHMAP, often had appended textual information (for example (CA:British Columbia)), it’s often good practice to first search using a wildcard character. For example, searching for “Marion” returns three sites:

marion_site <- get_site(sitename = 'Marion%')
print(marion_site)
##                          site.name       long      lat elev
##  Marion Lake (CA:British Columbia) -122.54722 49.30833  305
##  Marion Landfill                    -83.18611 40.59167   NA
##  Marion Lake                       -121.86241 44.55770 1259
## A site object containing 3 sites and 8 parameters.

You can also search by lat/lon bounding box. This one roughly corresponds to Florida.

FL_sites <- get_site(loc = c(-88, -79, 25, 30))

You can also search by geopolitical name or geopolitical IDs (gpid) stored in Neotoma. For a list of names and gpids, go to http://api.neotomadb.org/apdx/geopol.htm, or use the get_table(table.name = "GeoPoliticalUnits") command. This command works either with an explicit numeric ID, or with a text string:

# get all sites in New Mexico (gpid=7956)
NM_sites <- get_site(gpid = 7956)

# get all sites in Wisconsin
WI_sites <- get_site(gpid = "Wisconsin")

data.frame stores vectors of equal length. The nice thing about a data.frame is that each vector can be of a different type (character, numeric values, etc.). In RStudio, you can use the Environment panel in upper right to explore variables.

We pointed out before that the object returned from get_site is both a data.frame and a site object. Because it has a special print method some of the information from the full object is obscured when printed. You can see all the data in the data.frame using str (short for structure):

str(samwell_site)

Let’s look at the description field:

samwell_site$description

marion_site$description

4.0.2 Getting Datasets

The structure of the Neotoma data model, as expressed through the API is roughly: “counts within download, download within dataset, dataset within site”. So a dataset contains more information than a site, about a particular dataset from that site. A site may have a single associated dataset, or multiple. For example:

samwell_datasets <- get_dataset(samwell_site)

print(samwell_datasets)

get_dataset returns a list of datasets containing the metadata for each dataset

We can pass output from get_site to get_dataset, even if get_site returns multiple sites

marion.meta.dataset  <- get_dataset(marion_site)

Let’s look at the metadata returned for the Marion% search:

marion.meta.dataset

Both Marion Lake (CA: British Columbia) and Marion Landfill have a geochronology dataset, while Marion Lake (CA: British Columbia) has a pollen dataset and Marion Landfill a vertebrate fauna dataset. The third site, Marion Lake, has a diatom dataset and a water chemistry dataset.

Question 13: Are Marion Lake (CA: British Columbia) and Marion Lake the same site? Try using the browse function.

Further searches (i.e., to examine the lat/longs) or consulting the literature would be required. This illustrates the caution needed when using this, or any other, database. Human judgement still needs to be exercised when examining results from databases. The get_publication() function can return citation information if neccessary, but often using the browse() function is very helpful in examining a database further.

4.0.3 Get_Download

get_download() returns a list that stores a list of download objects - one for each retrieved dataset. Note that get_download() returns the actual data associated with each dataset, rather than a list of the available datasets, as in get_dataset() above. Each download object contains a suite of data for the samples in that dataset.

Get data for all datasets at Samwell Cave. get_download() will accept an object of class dataset (ie, samwell_dataset), but also of class site, since it will automatically query for the datasets associated in each site. Compare the returned messages for the following two commands:

samwell_all <- get_download(samwell_site)
## Warning in get_download.default(datasetid, verbose = verbose): Some datasetids returned empty downloads, be aware that length(datasetid) is longer than the download_list.
samwell_all <- get_download(samwell_datasets)
## Warning in get_download.default(datasetid, verbose = verbose): Some datasetids returned empty downloads, be aware that length(datasetid) is longer than the download_list.
print(samwell_all)

There are a number of messages that appear. These should be suppressed with the flag verbose = FALSE in the function call. One thing you’ll note is that not all of the datasets can be downloaded directly to a download objct. This is because geochronologic datasets have a different data structure than other data, requiring different fields, and as such, they can be obtained using the get_geochron() function:

samwell_geochron <- get_geochron(samwell_site)

print(samwell_geochron)

The result is effectively the inverse of the first.

Get the vertebrate datasets for just Samwell Cave Popcorn Dome (dataset 14262):

samwell_pd <- get_download(14262)
## API call was successful. Returned record for Samwell Cave

Let’s examine the available data in this download

str(samwell_pd[[1]])

There are 6 associated fields:

  1. dataset
    • site.data
    • dataset.meta
    • pi.data
    • submission
    • access.date
    • site
  2. sample.meta
  3. taxon.list
  4. counts
  5. lab.data
  6. chronologies

Within the download object, sample.meta stores the core depth and age information for that dataset. We just want to look at the first few lines, so are using the head() function. Let’s explore different facets of the dataset

head(samwell_pd[[1]]$sample.meta)

#taxon.list stores a list of taxa found  in the  dataset
head(samwell_pd[[1]]$taxon.list)

#counts stores the the counts, presence/absence data, or percentage data for each taxon for each sample
head(samwell_pd[[1]]$counts)

5 Multi-Site Analysis

If we have time, we can work through the example given the Neotoma package paper in Open Quaternary.

Jump to the section “Examples”, then scroll down to “Mammal Distributions in the Pleistocene”. The R code is reproduced below, with some modifications to accomodate changes since publication.

# install.packages('ggplot2','reshape2')
library("ggplot2")
library("reshape2") 
#  Bounding box is effectively the continental USA, excluding Alaska 
mam.set <- get_dataset(datasettype= 'vertebrate fauna', loc = c(-125, 30, -115, 49.5))

#  Retrieving this many sites can be very time consuming.
# If you are running code that takes a long time, but has an obvious output, you can stick it
# in an output file and call it later.  This is a nice trick :)

# if ("mam_dl.rds" %in% list.files('objects')) {
#   mam.dl <- readRDS('objects/mam_dl.rds')
# } else {
#   mam.dl <- get_download(mam.set)
#   saveRDS(mam.dl, file = 'objects/mam_dl.rds')
# }

githubURL <- "https://github.com/NeotomaDB/Workshops/blob/master/SVP2016/R/objects/mam.dl.RData?raw=true"
load(url(githubURL))

compiled.mam <- compile_downloads(mam.dl)

time.bins <- c(500, 4000, 10000, 15000, 20000) 

mean.age <- rowMeans(compiled.mam[,c('age.old','age.young', 'age')], na.rm = TRUE) 

interval <- findInterval(mean.age, time.bins) 

periods <- c('Modern',
             'Late Holocene',
             'Early-Mid Holocene',
             'Late Glacial',
             'Full Glacial',
             'Late Pleistocene') 

compiled.mam$ageInterval <- periods[interval + 1]

mam.melt <- melt(compiled.mam, measure.vars = 11:(ncol(compiled.mam)-1), na.rm = TRUE, factorsAsStrings = TRUE) 

mam.melt <- transform(mam.melt, ageInterval = factor(ageInterval, levels = periods)) 

mam.lat <- dcast(data = mam.melt, variable ~ageInterval, value.var = 'lat', 
                 fun.aggregate = mean, drop = TRUE)[,c(1, 3, 5, 6)]   # variable Late Holocene Late Glacial Full Glacial

# We only want taxa that appear at all time periods: 
mam.lat <- mam.lat[rowSums(is.na(mam.lat)) == 0, ] 

# Group the samples based on the range &direction (N vs S) of migration. 
# A shift of only 1 degree is considered stationary. 
mam.lat$grouping <- factor(findInterval (mam.lat[,2] - mam.lat[, 4], c(-11, -1, 1, 20)), 
                           labels = c('Southward','Stationary','Northward')) 
mam.lat.melt <- melt(mam.lat) 
colnames(mam.lat.melt)[2:3] <- c('cluster','Era')

ggplot(mam.lat.melt, aes(x = Era, y = value)) +
  geom_path(aes(group = variable, color =cluster)) +
  facet_wrap(~ cluster) + 
  scale_x_discrete(expand = c(.1,0)) + 
  ylab('Mean Latitude of Occurrence') +
  theme(axis.text.x = element_text(angle = 90, hjust = 1))

6 A few final thoughts

References

Goring, S., A. Dawson, G. L. Simpson, K. Ram, R. W. Graham, E. C. Grimm, and J. W. Williams. 2015. neotoma: A programmatic interface to the neotoma paleoecological database. Open Quaternary 1:2.

Funded by:
National Science Foundation
Directorate for Geosciences
Division of Earth Sciences 		EarthCube
Integrative Activities
National Sciences Foundation